Three ways to copy a molecule
In 1958 Matthew Meselson and Franklin Stahl set out to decide, among three published models, how a DNA molecule makes a copy of itself when a cell divides. James Watson and Francis Crick had proposed that the double helix splits into its two strands and each strand serves as the template for a new partner, so that every finished molecule carries one old strand and one newly made one, the semiconservative model. Rival proposals held that the whole original molecule survives intact while an entirely fresh copy is built alongside it, the conservative model, or that old and new DNA are broken and stitched together within both resulting strands, the dispersive model associated with Max Delbrück. Meselson and Stahl’s experiment was designed to tell these apart directly, by tracking the physical distribution of old and new DNA rather than inferring it.
Labelling DNA with heavy nitrogen
They grew Escherichia coli for many generations in a medium whose only nitrogen source was the heavier isotope nitrogen-15, so that every DNA molecule in the population became uniformly labelled with the heavy isotope. The bacteria were then transferred to ordinary medium containing the common isotope nitrogen-14, and samples were taken at intervals timed to successive rounds of division. DNA extracted from each sample was spun in a caesium chloride solution under high centrifugal force, a technique that separates molecules by small differences in density: heavier, nitrogen-15-rich DNA settles at one position in the tube, lighter nitrogen-14 DNA at another, and any hybrid molecule at a position in between. Comparing the resulting bands after each division cycle let them read off, directly, how old and new nitrogen were being combined into finished DNA.
Spinning DNA to see its density
The result matched only one of the three hypotheses. After a single round of division, all the DNA banded at a single density exactly midway between the heavy and light controls, a result the conservative model could not produce, since it predicted two separate bands, one fully heavy and one fully light, from the first division onward. After a second round, the DNA resolved into two bands in equal amounts: one at the intermediate density seen before, and one at the fully light density of unlabelled DNA. That pattern is precisely what semiconservative replication predicts, since each round distributes exactly one old strand into every new double helix, and it rules out the dispersive model, which would have kept producing a single band gradually approaching the light density rather than splitting cleanly in two.
One band, then two
The experiment settled the question of how the two strands are distributed once copying happens, but it was silent on how copying happens. It said nothing about the enzymes that unwind the double helix, synthesise the new strand, or correct errors as they occur, machinery that later research would spend decades characterising. It was also a single-organism demonstration: E. coli was well suited to nitrogen labelling and to dividing on a convenient schedule, and the experiment did not itself establish that eukaryotic cells, with their larger and more complex genomes, replicate DNA by the same pattern, even though that has since become the accepted general picture across life.
Ruling out the rivals
Semiconservative replication is the mechanism by which every cell hands a complete, functioning genome to each of its daughters, and by which an organism’s cells stay genetically consistent as they divide throughout a lifetime. Because each new double helix retains one original strand, replication also carries a built-in reference copy against which errors can, in principle, be checked, a feature that underlies the fidelity of inheritance across generations. The result gave molecular biology a settled starting assumption to build from: every technique that depends on DNA being copied strand for strand, from cloning to sequencing to diagnostic testing, rests on the picture that Meselson and Stahl’s bands first made visible.
What the experiment did not answer
This is a good one to sit with because the whole argument fits in a single figure: three predicted band patterns, one observed. There is no statistical argument to have and no ambiguous instrument reading; density either separates cleanly or it does not, and it did. Reading the original account is a compact lesson in how to design an experiment that eliminates rival explanations rather than merely supporting a favoured one, a discipline harder to find in messier fields of science. An hour with it repays anyone curious about how biology’s most basic housekeeping process was nailed down using little more than a growth medium and a centrifuge.